Artificial Intelligence Nanodegree

Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [1]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob

# define function to load train, test, and validation datasets
def load_dataset(path):
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('dogImages/train')
valid_files, valid_targets = load_dataset('dogImages/valid')
test_files, test_targets = load_dataset('dogImages/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("dogImages/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
Using TensorFlow backend.
/home/ec2-user/anaconda3/envs/tensorflow_p36/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
  return f(*args, **kwds)
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [2]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
There are 13233 total human images.

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[3])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 1

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [4]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer: based on cell below,

  • Detected 97.0 % of human faces in human_files_short
  • Detected 11.0 % of human faces in dog_files_short
In [5]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.

## TODO: Test the performance of the face_detector algorithm 
## on the images in human_files_short and dog_files_short.
faces_in_humans_percent = len([img_file for img_file in human_files_short if face_detector(img_file)])/len(human_files_short) *100
print("Detected {} % of human faces in human_files_short".format(faces_in_humans_percent))
faces_in_dogs_percent = len([img_file for img_file in dog_files_short if face_detector(img_file)])/len(dog_files_short) * 100
print("Detected {} % of human faces in dog_files_short".format(faces_in_dogs_percent))
Detected 97.0 % of human faces in human_files_short
Detected 11.0 % of human faces in dog_files_short

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer: Forcing the user to pose does not seems a reasonable expectation to me. The fact that this implementation not only look for the face but its position might be overkill for what we want if our need is just human face detection. We could use a CNN for this task, training on labeled dataset with a lot of different poses and noisy images. Another idea would be to add noise in the dataset, to make the presented face less clear.

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

In [6]:
## (Optional) TODO: Report the performance of another  
## face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.

Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.

In [7]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [8]:
from keras.preprocessing import image                  
from tqdm import tqdm

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [9]:
from keras.applications.resnet50 import preprocess_input, decode_predictions

def ResNet50_predict_labels(img_path):
    # returns prediction vector for image located at img_path
    img = preprocess_input(path_to_tensor(img_path))
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [10]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    prediction = ResNet50_predict_labels(img_path)
    return ((prediction <= 268) & (prediction >= 151)) 

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer: based on code cell below, this looks better ;-)

  • Detected 1.0 % of dogs in human_files_short
  • Detected 100.0 % of dogs in dog_files_short
In [11]:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.
dogs_in_humans_percent = len([img_file for img_file in human_files_short if dog_detector(img_file)])/len(human_files_short) *100
print("Detected {} % of dogs in human_files_short".format(dogs_in_humans_percent))
dogs_in_dogs_percent = len([img_file for img_file in dog_files_short if dog_detector(img_file)])/len(dog_files_short) * 100
print("Detected {} % of dogs in dog_files_short".format(dogs_in_dogs_percent))
Detected 1.0 % of dogs in human_files_short
Detected 100.0 % of dogs in dog_files_short

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.

In [12]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255
100%|██████████| 6680/6680 [00:54<00:00, 123.53it/s]
100%|██████████| 835/835 [00:06<00:00, 159.09it/s]
100%|██████████| 836/836 [00:05<00:00, 168.49it/s]

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer: I started with the architecture from the tutorial on MNIST, just adapting the input shape and the output layer with softmax for the 133 breed. this gave me around 6% with 5 epochs on my cpu. I then tried data augmentation and more epoch on gpu to reach ~18%. I also use athe ws gpu instances to experiment with different epoch/batch size configurations

In [71]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Activation, Dropout, Flatten, Dense, BatchNormalization
from keras.models import Sequential

model = Sequential()

### TODO: Define your architecture.
model.add(Conv2D(filters=16, kernel_size=2, padding='same', input_shape=train_tensors.shape[1:], activation='relu'))
model.add(MaxPooling2D(pool_size=2))
          
model.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=2))     
          
model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=2))      

model.add(Flatten())        
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(133, activation='softmax'))

model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_82 (Conv2D)           (None, 224, 224, 16)      208       
_________________________________________________________________
max_pooling2d_11 (MaxPooling (None, 112, 112, 16)      0         
_________________________________________________________________
conv2d_83 (Conv2D)           (None, 112, 112, 32)      2080      
_________________________________________________________________
max_pooling2d_12 (MaxPooling (None, 56, 56, 32)        0         
_________________________________________________________________
conv2d_84 (Conv2D)           (None, 56, 56, 64)        8256      
_________________________________________________________________
max_pooling2d_13 (MaxPooling (None, 28, 28, 64)        0         
_________________________________________________________________
flatten_5 (Flatten)          (None, 50176)             0         
_________________________________________________________________
dropout_7 (Dropout)          (None, 50176)             0         
_________________________________________________________________
dense_13 (Dense)             (None, 512)               25690624  
_________________________________________________________________
dropout_8 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_14 (Dense)             (None, 133)               68229     
=================================================================
Total params: 25,769,397.0
Trainable params: 25,769,397.0
Non-trainable params: 0.0
_________________________________________________________________

Compile the Model

In [72]:
model.compile(optimizer='nadam', loss='categorical_crossentropy', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [73]:
from keras.callbacks import ModelCheckpoint
from keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img 

### TODO: specify the number of epochs that you would like to use to train the model.

epochs = 5
batch_size = 128
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', 
                               verbose=1, save_best_only=True)

%time model.fit(train_tensors, train_targets, \
          validation_data=(valid_tensors, valid_targets), \
          epochs=epochs, batch_size=batch_size, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/5
6656/6680 [============================>.] - ETA: 0s - loss: 5.2044 - acc: 0.0140Epoch 00000: val_loss improved from inf to 4.74050, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 14s - loss: 5.2028 - acc: 0.0139 - val_loss: 4.7405 - val_acc: 0.0299
Epoch 2/5
6656/6680 [============================>.] - ETA: 0s - loss: 4.5354 - acc: 0.0391Epoch 00001: val_loss improved from 4.74050 to 4.43649, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 6s - loss: 4.5354 - acc: 0.0391 - val_loss: 4.4365 - val_acc: 0.0479
Epoch 3/5
6656/6680 [============================>.] - ETA: 0s - loss: 4.1668 - acc: 0.0798Epoch 00002: val_loss did not improve
6680/6680 [==============================] - 6s - loss: 4.1661 - acc: 0.0796 - val_loss: 4.9401 - val_acc: 0.0323
Epoch 4/5
6656/6680 [============================>.] - ETA: 0s - loss: 3.7787 - acc: 0.1435Epoch 00003: val_loss improved from 4.43649 to 4.23504, saving model to saved_models/weights.best.from_scratch.hdf5
6680/6680 [==============================] - 6s - loss: 3.7774 - acc: 0.1437 - val_loss: 4.2350 - val_acc: 0.0814
Epoch 5/5
6656/6680 [============================>.] - ETA: 0s - loss: 2.9248 - acc: 0.3000Epoch 00004: val_loss did not improve
6680/6680 [==============================] - 6s - loss: 2.9262 - acc: 0.2996 - val_loss: 4.6230 - val_acc: 0.0743
CPU times: user 26.8 s, sys: 11.6 s, total: 38.4 s
Wall time: 42 s
Out[73]:
<keras.callbacks.History at 0x7fcd415b4f28>
In [76]:
##trying data augmentation on the fly using loaded tensors
epochs = 20
batch_size = 128
train_datagen = ImageDataGenerator(
        featurewise_center=False,
        samplewise_center=False,
        featurewise_std_normalization=False,
        samplewise_std_normalization=False,
        zca_whitening=False,
        rotation_range=45,
        width_shift_range=.1,
        height_shift_range=.1,
        shear_range=0.3,
        zoom_range=0.3,
        channel_shift_range=0,
        fill_mode='nearest',
        cval=0.,
        horizontal_flip=True,
        vertical_flip=False)

valid_datagen = ImageDataGenerator()

train_generator = train_datagen.flow(
        train_tensors,
        train_targets,
        batch_size=batch_size)

validation_generator = valid_datagen.flow(
        valid_tensors, 
        valid_targets,
        batch_size=batch_size)

checkpointerDataAugmented = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch_augmented.hdf5', 
                               verbose=1, save_best_only=True)

%time model.fit_generator( \
        train_generator, \
        steps_per_epoch=train_tensors.shape[0]//batch_size, \
        epochs=epochs, \
        validation_data=validation_generator, \
        validation_steps=valid_tensors.shape[0]//batch_size, \
        callbacks=[checkpointerDataAugmented], verbose=1)
Epoch 1/20
51/52 [============================>.] - ETA: 1s - loss: 4.2131 - acc: 0.0699Epoch 00000: val_loss improved from inf to 4.06243, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 53s - loss: 4.2099 - acc: 0.0702 - val_loss: 4.0624 - val_acc: 0.0924
Epoch 2/20
51/52 [============================>.] - ETA: 0s - loss: 4.1230 - acc: 0.0833Epoch 00001: val_loss improved from 4.06243 to 3.94783, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 4.1214 - acc: 0.0831 - val_loss: 3.9478 - val_acc: 0.1117
Epoch 3/20
51/52 [============================>.] - ETA: 0s - loss: 4.1390 - acc: 0.0775Epoch 00002: val_loss did not improve
52/52 [==============================] - 51s - loss: 4.1389 - acc: 0.0777 - val_loss: 4.0031 - val_acc: 0.1033
Epoch 4/20
51/52 [============================>.] - ETA: 0s - loss: 3.9806 - acc: 0.0994Epoch 00003: val_loss improved from 3.94783 to 3.92616, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.9806 - acc: 0.0992 - val_loss: 3.9262 - val_acc: 0.1231
Epoch 5/20
51/52 [============================>.] - ETA: 0s - loss: 3.9424 - acc: 0.0992Epoch 00004: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.9418 - acc: 0.0993 - val_loss: 3.9331 - val_acc: 0.1160
Epoch 6/20
51/52 [============================>.] - ETA: 0s - loss: 3.8690 - acc: 0.1073Epoch 00005: val_loss improved from 3.92616 to 3.87484, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.8697 - acc: 0.1075 - val_loss: 3.8748 - val_acc: 0.1259
Epoch 7/20
51/52 [============================>.] - ETA: 0s - loss: 3.8271 - acc: 0.1079Epoch 00006: val_loss improved from 3.87484 to 3.83541, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.8328 - acc: 0.1069 - val_loss: 3.8354 - val_acc: 0.1202
Epoch 8/20
51/52 [============================>.] - ETA: 0s - loss: 3.7668 - acc: 0.1254Epoch 00007: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.7681 - acc: 0.1250 - val_loss: 3.8543 - val_acc: 0.1216
Epoch 9/20
51/52 [============================>.] - ETA: 0s - loss: 3.7400 - acc: 0.1286Epoch 00008: val_loss improved from 3.83541 to 3.83176, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.7404 - acc: 0.1277 - val_loss: 3.8318 - val_acc: 0.1245
Epoch 10/20
51/52 [============================>.] - ETA: 0s - loss: 3.6798 - acc: 0.1363Epoch 00009: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.6865 - acc: 0.1361 - val_loss: 3.8399 - val_acc: 0.1188
Epoch 11/20
51/52 [============================>.] - ETA: 0s - loss: 3.6178 - acc: 0.1448Epoch 00010: val_loss improved from 3.83176 to 3.75618, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.6187 - acc: 0.1454 - val_loss: 3.7562 - val_acc: 0.1202
Epoch 12/20
51/52 [============================>.] - ETA: 0s - loss: 3.5584 - acc: 0.1479Epoch 00011: val_loss improved from 3.75618 to 3.68821, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.5587 - acc: 0.1476 - val_loss: 3.6882 - val_acc: 0.1386
Epoch 13/20
51/52 [============================>.] - ETA: 0s - loss: 3.5763 - acc: 0.1436Epoch 00012: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.5716 - acc: 0.1443 - val_loss: 3.6995 - val_acc: 0.1471
Epoch 14/20
51/52 [============================>.] - ETA: 0s - loss: 3.4638 - acc: 0.1694Epoch 00013: val_loss improved from 3.68821 to 3.65104, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.4635 - acc: 0.1690 - val_loss: 3.6510 - val_acc: 0.1471
Epoch 15/20
51/52 [============================>.] - ETA: 0s - loss: 3.4510 - acc: 0.1674Epoch 00014: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.4506 - acc: 0.1664 - val_loss: 3.6705 - val_acc: 0.1429
Epoch 16/20
51/52 [============================>.] - ETA: 0s - loss: 3.4330 - acc: 0.1674Epoch 00015: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.4324 - acc: 0.1681 - val_loss: 3.6514 - val_acc: 0.1641
Epoch 17/20
51/52 [============================>.] - ETA: 0s - loss: 3.3843 - acc: 0.1723Epoch 00016: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.3812 - acc: 0.1732 - val_loss: 3.6946 - val_acc: 0.1429
Epoch 18/20
51/52 [============================>.] - ETA: 0s - loss: 3.3641 - acc: 0.1867Epoch 00017: val_loss improved from 3.65104 to 3.62811, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.3626 - acc: 0.1866 - val_loss: 3.6281 - val_acc: 0.1711
Epoch 19/20
51/52 [============================>.] - ETA: 0s - loss: 3.3165 - acc: 0.1951Epoch 00018: val_loss improved from 3.62811 to 3.60463, saving model to saved_models/weights.best.from_scratch_augmented.hdf5
52/52 [==============================] - 52s - loss: 3.3154 - acc: 0.1951 - val_loss: 3.6046 - val_acc: 0.1612
Epoch 20/20
51/52 [============================>.] - ETA: 0s - loss: 3.2673 - acc: 0.2010Epoch 00019: val_loss did not improve
52/52 [==============================] - 51s - loss: 3.2679 - acc: 0.2004 - val_loss: 3.6474 - val_acc: 0.1796
CPU times: user 17min 3s, sys: 1min 37s, total: 18min 40s
Wall time: 17min 23s
Out[76]:
<keras.callbacks.History at 0x7fcd41569940>

Load the Model with the Best Validation Loss

In [77]:
model.load_weights('saved_models/weights.best.from_scratch_augmented.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [78]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 18.8995%

Step 4: Use a CNN to Classify Dog Breeds

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features

In [19]:
bottleneck_features = np.load('bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']

Model Architecture

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.

In [20]:
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
VGG16_model.add(Dense(133, activation='softmax'))

VGG16_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_1 ( (None, 512)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 133)               68229     
=================================================================
Total params: 68,229.0
Trainable params: 68,229.0
Non-trainable params: 0.0
_________________________________________________________________

Compile the Model

In [21]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

Train the Model

In [22]:
from keras.callbacks import ModelCheckpoint  
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                               verbose=1, save_best_only=True)

VGG16_model.fit(train_VGG16, train_targets, 
          validation_data=(valid_VGG16, valid_targets),
          epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6580/6680 [============================>.] - ETA: 0s - loss: 12.3377 - acc: 0.1287Epoch 00000: val_loss improved from inf to 10.78259, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 12.3232 - acc: 0.1299 - val_loss: 10.7826 - val_acc: 0.2251
Epoch 2/20
6600/6680 [============================>.] - ETA: 0s - loss: 10.3555 - acc: 0.2770Epoch 00001: val_loss improved from 10.78259 to 10.26565, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 10.3479 - acc: 0.2769 - val_loss: 10.2657 - val_acc: 0.2754
Epoch 3/20
6480/6680 [============================>.] - ETA: 0s - loss: 9.8719 - acc: 0.3373Epoch 00002: val_loss improved from 10.26565 to 10.02771, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.8876 - acc: 0.3365 - val_loss: 10.0277 - val_acc: 0.3114
Epoch 4/20
6620/6680 [============================>.] - ETA: 0s - loss: 9.7468 - acc: 0.3624Epoch 00003: val_loss improved from 10.02771 to 10.01347, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.7644 - acc: 0.3615 - val_loss: 10.0135 - val_acc: 0.3281
Epoch 5/20
6620/6680 [============================>.] - ETA: 0s - loss: 9.6298 - acc: 0.3776Epoch 00004: val_loss improved from 10.01347 to 9.87191, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.6342 - acc: 0.3774 - val_loss: 9.8719 - val_acc: 0.3401
Epoch 6/20
6540/6680 [============================>.] - ETA: 0s - loss: 9.5021 - acc: 0.3890Epoch 00005: val_loss improved from 9.87191 to 9.78245, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.5037 - acc: 0.3888 - val_loss: 9.7824 - val_acc: 0.3257
Epoch 7/20
6620/6680 [============================>.] - ETA: 0s - loss: 9.3436 - acc: 0.4012Epoch 00006: val_loss improved from 9.78245 to 9.71230, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.3386 - acc: 0.4012 - val_loss: 9.7123 - val_acc: 0.3413
Epoch 8/20
6460/6680 [============================>.] - ETA: 0s - loss: 9.2929 - acc: 0.4111Epoch 00007: val_loss improved from 9.71230 to 9.67956, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.2969 - acc: 0.4109 - val_loss: 9.6796 - val_acc: 0.3437
Epoch 9/20
6580/6680 [============================>.] - ETA: 0s - loss: 9.2189 - acc: 0.4128Epoch 00008: val_loss improved from 9.67956 to 9.52508, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.2080 - acc: 0.4133 - val_loss: 9.5251 - val_acc: 0.3677
Epoch 10/20
6600/6680 [============================>.] - ETA: 0s - loss: 9.0416 - acc: 0.4277Epoch 00009: val_loss improved from 9.52508 to 9.39770, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 9.0522 - acc: 0.4271 - val_loss: 9.3977 - val_acc: 0.3725
Epoch 11/20
6600/6680 [============================>.] - ETA: 0s - loss: 8.9804 - acc: 0.4330Epoch 00010: val_loss improved from 9.39770 to 9.37424, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.9887 - acc: 0.4325 - val_loss: 9.3742 - val_acc: 0.3701
Epoch 12/20
6540/6680 [============================>.] - ETA: 0s - loss: 8.8665 - acc: 0.4341Epoch 00011: val_loss improved from 9.37424 to 9.20944, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.8597 - acc: 0.4341 - val_loss: 9.2094 - val_acc: 0.3760
Epoch 13/20
6460/6680 [============================>.] - ETA: 0s - loss: 8.7118 - acc: 0.4426Epoch 00012: val_loss improved from 9.20944 to 8.99589, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.7209 - acc: 0.4422 - val_loss: 8.9959 - val_acc: 0.3856
Epoch 14/20
6400/6680 [===========================>..] - ETA: 0s - loss: 8.5907 - acc: 0.4497Epoch 00013: val_loss improved from 8.99589 to 8.91744, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.5820 - acc: 0.4496 - val_loss: 8.9174 - val_acc: 0.3820
Epoch 15/20
6640/6680 [============================>.] - ETA: 0s - loss: 8.3459 - acc: 0.4646Epoch 00014: val_loss improved from 8.91744 to 8.82339, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.3374 - acc: 0.4651 - val_loss: 8.8234 - val_acc: 0.3940
Epoch 16/20
6640/6680 [============================>.] - ETA: 0s - loss: 8.2893 - acc: 0.4747Epoch 00015: val_loss improved from 8.82339 to 8.73113, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.2783 - acc: 0.4754 - val_loss: 8.7311 - val_acc: 0.4144
Epoch 17/20
6600/6680 [============================>.] - ETA: 0s - loss: 8.2041 - acc: 0.4827Epoch 00016: val_loss improved from 8.73113 to 8.60047, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.2120 - acc: 0.4823 - val_loss: 8.6005 - val_acc: 0.4156
Epoch 18/20
6580/6680 [============================>.] - ETA: 0s - loss: 8.0602 - acc: 0.4904Epoch 00017: val_loss improved from 8.60047 to 8.58259, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 8.0717 - acc: 0.4894 - val_loss: 8.5826 - val_acc: 0.4144
Epoch 19/20
6560/6680 [============================>.] - ETA: 0s - loss: 7.9808 - acc: 0.4939Epoch 00018: val_loss improved from 8.58259 to 8.50988, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 7.9629 - acc: 0.4952 - val_loss: 8.5099 - val_acc: 0.4108
Epoch 20/20
6560/6680 [============================>.] - ETA: 0s - loss: 7.9078 - acc: 0.5018Epoch 00019: val_loss improved from 8.50988 to 8.43930, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s - loss: 7.8913 - acc: 0.5030 - val_loss: 8.4393 - val_acc: 0.4263
Out[22]:
<keras.callbacks.History at 0x7fcff472be80>

Load the Model with the Best Validation Loss

In [23]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')

Test the Model

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.

In [24]:
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]

# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 40.4306%

Predict Dog Breed with the Model

In [25]:
from extract_bottleneck_features import *

def VGG16_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = VGG16_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras:

The files are encoded as such:

Dog{network}Data.npz

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception. Pick one of the above architectures, download the corresponding bottleneck features, and store the downloaded file in the bottleneck_features/ folder in the repository.

(IMPLEMENTATION) Obtain Bottleneck Features

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']
In [26]:
### TODO: Obtain bottleneck features from another pre-trained CNN.
bottleneck_features_Xception = np.load('bottleneck_features/DogXceptionData.npz')
train_Xception = bottleneck_features_Xception['train']
valid_Xception = bottleneck_features_Xception['valid']
test_Xception = bottleneck_features_Xception['test']

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer: I started with VGG_19 and simple GlobalAveragePooling2D followed by a dense layer. this achieved 57 % accuracy on the test set with 20 epochs, I then added a Dense layer with 500 nodes in beetween with a relu activation, and with dropouts, this achived 71 % accuracy. I then tried all available bottleneck features to try same architecture on all pre-trained network. From there I chose Xception that yielded the best precision ~83%.I then decided to simplify and only connect a dense 133 layer with a softmax, this gave me the best output ~86% accuracy for the fastet training, even on cpu !

In [51]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense
from keras.models import Sequential
myhotdog_model = Sequential()

myhotdog_model.add(GlobalAveragePooling2D(input_shape=train_Xception.shape[1:]))
myhotdog_model.add(Dense(133, activation='softmax'))
myhotdog_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_6 ( (None, 2048)              0         
_________________________________________________________________
dense_8 (Dense)              (None, 133)               272517    
=================================================================
Total params: 272,517.0
Trainable params: 272,517.0
Non-trainable params: 0.0
_________________________________________________________________

(IMPLEMENTATION) Compile the Model

In [52]:
### TODO: Compile the model.
myhotdog_model.compile(loss='categorical_crossentropy', optimizer='nadam', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [53]:
from keras.callbacks import ModelCheckpoint 

### TODO: Train the model.
checkpointer_myhotdog = ModelCheckpoint(filepath='saved_models/weights.best.myhotdog.hdf5', 
                               verbose=1, save_best_only=True)

%time myhotdog_model.fit(train_Xception, train_targets, \
          validation_data=(valid_Xception, valid_targets), \
          epochs=50, batch_size=2048, callbacks=[checkpointer_myhotdog], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/50
6144/6680 [==========================>...] - ETA: 0s - loss: 4.3005 - acc: 0.2106Epoch 00000: val_loss improved from inf to 2.88943, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 4s - loss: 4.2134 - acc: 0.2418 - val_loss: 2.8894 - val_acc: 0.6144
Epoch 2/50
6144/6680 [==========================>...] - ETA: 0s - loss: 2.5064 - acc: 0.6911Epoch 00001: val_loss improved from 2.88943 to 1.74723, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 2.4595 - acc: 0.6976 - val_loss: 1.7472 - val_acc: 0.7317
Epoch 3/50
6144/6680 [==========================>...] - ETA: 0s - loss: 1.4814 - acc: 0.8011Epoch 00002: val_loss improved from 1.74723 to 1.13682, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 1.4609 - acc: 0.8000 - val_loss: 1.1368 - val_acc: 0.7952
Epoch 4/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.9541 - acc: 0.8418Epoch 00003: val_loss improved from 1.13682 to 0.84506, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.9491 - acc: 0.8412 - val_loss: 0.8451 - val_acc: 0.8216
Epoch 5/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.7131 - acc: 0.8605Epoch 00004: val_loss improved from 0.84506 to 0.69697, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.7040 - acc: 0.8626 - val_loss: 0.6970 - val_acc: 0.8419
Epoch 6/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.5779 - acc: 0.8787Epoch 00005: val_loss improved from 0.69697 to 0.61514, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.5692 - acc: 0.8816 - val_loss: 0.6151 - val_acc: 0.8395
Epoch 7/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.4873 - acc: 0.8877Epoch 00006: val_loss improved from 0.61514 to 0.56746, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.4865 - acc: 0.8882 - val_loss: 0.5675 - val_acc: 0.8539
Epoch 8/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.4381 - acc: 0.8983Epoch 00007: val_loss improved from 0.56746 to 0.54009, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.4300 - acc: 0.9004 - val_loss: 0.5401 - val_acc: 0.8551
Epoch 9/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.3891 - acc: 0.9098Epoch 00008: val_loss improved from 0.54009 to 0.51350, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.3887 - acc: 0.9100 - val_loss: 0.5135 - val_acc: 0.8527
Epoch 10/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.3564 - acc: 0.9150Epoch 00009: val_loss improved from 0.51350 to 0.50618, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.3559 - acc: 0.9156 - val_loss: 0.5062 - val_acc: 0.8503
Epoch 11/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.3281 - acc: 0.9212Epoch 00010: val_loss improved from 0.50618 to 0.49432, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.3274 - acc: 0.9213 - val_loss: 0.4943 - val_acc: 0.8551
Epoch 12/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.3059 - acc: 0.9271Epoch 00011: val_loss improved from 0.49432 to 0.47630, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.3061 - acc: 0.9269 - val_loss: 0.4763 - val_acc: 0.8575
Epoch 13/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.2818 - acc: 0.9342Epoch 00012: val_loss improved from 0.47630 to 0.46290, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.2832 - acc: 0.9344 - val_loss: 0.4629 - val_acc: 0.8551
Epoch 14/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.2619 - acc: 0.9412Epoch 00013: val_loss improved from 0.46290 to 0.45864, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.2638 - acc: 0.9401 - val_loss: 0.4586 - val_acc: 0.8599
Epoch 15/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.2492 - acc: 0.9451Epoch 00014: val_loss improved from 0.45864 to 0.45123, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.2488 - acc: 0.9454 - val_loss: 0.4512 - val_acc: 0.8635
Epoch 16/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.2342 - acc: 0.9507Epoch 00015: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.2337 - acc: 0.9504 - val_loss: 0.4555 - val_acc: 0.8563
Epoch 17/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.2208 - acc: 0.9518Epoch 00016: val_loss improved from 0.45123 to 0.44482, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.2208 - acc: 0.9518 - val_loss: 0.4448 - val_acc: 0.8575
Epoch 18/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.2090 - acc: 0.9574Epoch 00017: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.2073 - acc: 0.9572 - val_loss: 0.4468 - val_acc: 0.8611
Epoch 19/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1949 - acc: 0.9629Epoch 00018: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1972 - acc: 0.9606 - val_loss: 0.4485 - val_acc: 0.8551
Epoch 20/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1892 - acc: 0.9642Epoch 00019: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1885 - acc: 0.9645 - val_loss: 0.4463 - val_acc: 0.8551
Epoch 21/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1769 - acc: 0.9683Epoch 00020: val_loss improved from 0.44482 to 0.43783, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.1778 - acc: 0.9680 - val_loss: 0.4378 - val_acc: 0.8635
Epoch 22/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1695 - acc: 0.9722Epoch 00021: val_loss improved from 0.43783 to 0.43517, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.1687 - acc: 0.9723 - val_loss: 0.4352 - val_acc: 0.8623
Epoch 23/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1616 - acc: 0.9748Epoch 00022: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1614 - acc: 0.9749 - val_loss: 0.4377 - val_acc: 0.8671
Epoch 24/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1542 - acc: 0.9771Epoch 00023: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1534 - acc: 0.9771 - val_loss: 0.4354 - val_acc: 0.8575
Epoch 25/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1445 - acc: 0.9803Epoch 00024: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1459 - acc: 0.9798 - val_loss: 0.4354 - val_acc: 0.8611
Epoch 26/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1393 - acc: 0.9808Epoch 00025: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1401 - acc: 0.9807 - val_loss: 0.4364 - val_acc: 0.8623
Epoch 27/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1342 - acc: 0.9824Epoch 00026: val_loss improved from 0.43517 to 0.42756, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.1339 - acc: 0.9823 - val_loss: 0.4276 - val_acc: 0.8611
Epoch 28/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1302 - acc: 0.9836Epoch 00027: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1288 - acc: 0.9838 - val_loss: 0.4306 - val_acc: 0.8563
Epoch 29/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1247 - acc: 0.9834Epoch 00028: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1237 - acc: 0.9840 - val_loss: 0.4284 - val_acc: 0.8659
Epoch 30/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1173 - acc: 0.9876Epoch 00029: val_loss improved from 0.42756 to 0.42678, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.1190 - acc: 0.9867 - val_loss: 0.4268 - val_acc: 0.8647
Epoch 31/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1149 - acc: 0.9875Epoch 00030: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1145 - acc: 0.9873 - val_loss: 0.4339 - val_acc: 0.8575
Epoch 32/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1068 - acc: 0.9893Epoch 00031: val_loss improved from 0.42678 to 0.42355, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.1093 - acc: 0.9885 - val_loss: 0.4236 - val_acc: 0.8623
Epoch 33/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1062 - acc: 0.9886Epoch 00032: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1057 - acc: 0.9888 - val_loss: 0.4262 - val_acc: 0.8635
Epoch 34/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.1023 - acc: 0.9894Epoch 00033: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.1015 - acc: 0.9898 - val_loss: 0.4338 - val_acc: 0.8575
Epoch 35/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0987 - acc: 0.9899Epoch 00034: val_loss improved from 0.42355 to 0.42267, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.0980 - acc: 0.9903 - val_loss: 0.4227 - val_acc: 0.8659
Epoch 36/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0952 - acc: 0.9912Epoch 00035: val_loss improved from 0.42267 to 0.42192, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.0948 - acc: 0.9915 - val_loss: 0.4219 - val_acc: 0.8635
Epoch 37/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0918 - acc: 0.9907Epoch 00036: val_loss improved from 0.42192 to 0.41818, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.0916 - acc: 0.9909 - val_loss: 0.4182 - val_acc: 0.8659
Epoch 38/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0897 - acc: 0.9917Epoch 00037: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0890 - acc: 0.9919 - val_loss: 0.4208 - val_acc: 0.8683
Epoch 39/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0855 - acc: 0.9937Epoch 00038: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0855 - acc: 0.9933 - val_loss: 0.4217 - val_acc: 0.8635
Epoch 40/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0827 - acc: 0.9935Epoch 00039: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0832 - acc: 0.9936 - val_loss: 0.4235 - val_acc: 0.8659
Epoch 41/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0811 - acc: 0.9928Epoch 00040: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0807 - acc: 0.9930 - val_loss: 0.4226 - val_acc: 0.8623
Epoch 42/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0774 - acc: 0.9951Epoch 00041: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0778 - acc: 0.9945 - val_loss: 0.4231 - val_acc: 0.8659
Epoch 43/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0763 - acc: 0.9938Epoch 00042: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0758 - acc: 0.9942 - val_loss: 0.4206 - val_acc: 0.8623
Epoch 44/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0733 - acc: 0.9951Epoch 00043: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0732 - acc: 0.9949 - val_loss: 0.4235 - val_acc: 0.8611
Epoch 45/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0718 - acc: 0.9950Epoch 00044: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0714 - acc: 0.9952 - val_loss: 0.4269 - val_acc: 0.8623
Epoch 46/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0697 - acc: 0.9959Epoch 00045: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0695 - acc: 0.9960 - val_loss: 0.4252 - val_acc: 0.8635
Epoch 47/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0671 - acc: 0.9961Epoch 00046: val_loss improved from 0.41818 to 0.41731, saving model to saved_models/weights.best.myhotdog.hdf5
6680/6680 [==============================] - 1s - loss: 0.0675 - acc: 0.9958 - val_loss: 0.4173 - val_acc: 0.8671
Epoch 48/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0651 - acc: 0.9969Epoch 00047: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0655 - acc: 0.9969 - val_loss: 0.4256 - val_acc: 0.8683
Epoch 49/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0629 - acc: 0.9963Epoch 00048: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0638 - acc: 0.9963 - val_loss: 0.4210 - val_acc: 0.8743
Epoch 50/50
6144/6680 [==========================>...] - ETA: 0s - loss: 0.0616 - acc: 0.9956Epoch 00049: val_loss did not improve
6680/6680 [==============================] - 1s - loss: 0.0620 - acc: 0.9957 - val_loss: 0.4242 - val_acc: 0.8623
CPU times: user 41.5 s, sys: 1min 1s, total: 1min 42s
Wall time: 1min 38s
Out[53]:
<keras.callbacks.History at 0x7fcd64375240>

(IMPLEMENTATION) Load the Model with the Best Validation Loss

In [54]:
### TODO: Load the model weights with the best validation loss.
myhotdog_model.load_weights('saved_models/weights.best.myhotdog.hdf5')

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.

In [55]:
### TODO: Calculate classification accuracy on the test dataset.
# get index of predicted dog breed for each image in test set
myhotdog_model.load_weights('saved_models/weights.best.myhotdog.hdf5')
myhotdog_predictions = [np.argmax(myhotdog_model.predict(np.expand_dims(feature, axis=0))) for feature in test_Xception]
myhotdog_test_accuracy = 100*np.sum(np.array(myhotdog_predictions)==np.argmax(test_targets, axis=1))/len(myhotdog_predictions)
print('myhotdog Test accuracy: %.4f%%' % myhotdog_test_accuracy)
myhotdog Test accuracy: 86.9617%

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 5, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.

In [56]:
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.
from extract_bottleneck_features import *

def Myhotdog_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_Xception(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = myhotdog_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm

In [57]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.
%matplotlib inline   
import numpy as np
import os
import random

from IPython.display import display, Image

from glob import glob

  

def dogface(img_path):
    #start with dog_detector as it is more accurate on non-dog image
    category = ""
    if (dog_detector(img_path)):
        category = "dog"
    else:
        if face_detector(img_path):
            category = "human"
        else: 
            category = "not dog, not human"
            print("ERROR: could not recognize a human or a dog, will still try to guess...")
            
    breed=Myhotdog_predict_breed(img_path)
    welcome="Woof to you, {} !".format(category)
    guess="seems to me you look like a {}".format(breed)
    print(welcome)
    display(Image(filename=img_path, width=420))
    print(guess)
    #find a random img for this breed in training data
    breedImages = "dogImages/train/*{}*/*.jpg".format(breed)
    display(Image(filename=random.choice(glob(breedImages)), width=420))
    
        
    

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer: I like the output, it was actually better than what I expected ! I saw a difference with the pictures , moving from 71% accuracy to almost 87% accuracy made corrected some predicitions: the Bernese_mountain_dog was always classified as a cavalier king charles spaniel in the earliest version I think I could work on the following to improve the algorithm:

  • data augmentation for transfer learning: I got stuck here and did not manage to successfully generate the features with augmentation (using Keras generator and model.predict ... but always failed with memory error)
  • maybe pre-process the image in input to focus only on the face , this way we would compare human face with dog face, rather than the variety of human photo poses to the variety of dog photo poses...
  • maybe return 2 solutions when the probabilities from the softmax are close, this way we could increase the chance of getting the breed right for mixed breed...
  • try to return the closest image match from the training sample (right now I'm returning a random one)
In [58]:
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.

test_files = np.array(glob("dogfaceImages/*"))
for img in test_files:
    print("Running DogFace on {}".format(img))
    %time dogface(img)
    
Running DogFace on dogfaceImages/leonar.jpg
ERROR: could not recognize a human or a dog, will still try to guess...
Woof to you, not dog, not human !
seems to me you look like a American_foxhound
CPU times: user 10.5 s, sys: 36 ms, total: 10.6 s
Wall time: 9.87 s
Running DogFace on dogfaceImages/lu.jpg
Woof to you, human !
seems to me you look like a Dachshund
CPU times: user 11.8 s, sys: 8 ms, total: 11.8 s
Wall time: 11.4 s
Running DogFace on dogfaceImages/me.jpg
Woof to you, human !
seems to me you look like a Black_and_tan_coonhound
CPU times: user 11.3 s, sys: 28 ms, total: 11.4 s
Wall time: 10.6 s
Running DogFace on dogfaceImages/dog2.JPG
Woof to you, dog !
seems to me you look like a Canaan_dog
CPU times: user 10.6 s, sys: 8 ms, total: 10.6 s
Wall time: 10.4 s
Running DogFace on dogfaceImages/dog5.JPG
Woof to you, dog !
seems to me you look like a Cavalier_king_charles_spaniel
CPU times: user 12.2 s, sys: 0 ns, total: 12.2 s
Wall time: 12 s
Running DogFace on dogfaceImages/dog4.JPG
Woof to you, dog !
seems to me you look like a Bernese_mountain_dog
CPU times: user 11.2 s, sys: 56 ms, total: 11.2 s
Wall time: 11.1 s
Running DogFace on dogfaceImages/dog1.JPG
Woof to you, dog !
seems to me you look like a Anatolian_shepherd_dog
CPU times: user 13 s, sys: 20 ms, total: 13 s
Wall time: 12.9 s
Running DogFace on dogfaceImages/dog3.JPG
Woof to you, dog !
seems to me you look like a Border_collie
CPU times: user 11.5 s, sys: 100 ms, total: 11.6 s
Wall time: 11.5 s
Running DogFace on dogfaceImages/dog6.jpg
ERROR: could not recognize a human or a dog, will still try to guess...
Woof to you, not dog, not human !
seems to me you look like a Smooth_fox_terrier
CPU times: user 11.9 s, sys: 28 ms, total: 11.9 s
Wall time: 11.6 s